sys/amd64/amd64/trap.c

   1 /*-
   2  * Copyright (C) 1994, David Greenman
   3  * Copyright (c) 1990, 1993
   4  *      The Regents of the University of California.  All rights reserved.
   5  *
   6  * This code is derived from software contributed to Berkeley by
   7  * the University of Utah, and William Jolitz.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  * 3. All advertising materials mentioning features or use of this software
  18  *    must display the following acknowledgement:
  19  *      This product includes software developed by the University of
  20  *      California, Berkeley and its contributors.
  21  * 4. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      from: @(#)trap.c        7.4 (Berkeley) 5/13/91
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 /*
  44  * AMD64 Trap and System call handling
  45  */
  46
  47 #include "opt_clock.h"
  48 #include "opt_cpu.h"
  49 #include "opt_hwpmc_hooks.h"
  50 #include "opt_isa.h"
  51 #include "opt_kdb.h"
  52 #include "opt_kdtrace.h"
  53
  54 #include <sys/param.h>
  55 #include <sys/bus.h>
  56 #include <sys/systm.h>
  57 #include <sys/proc.h>
  58 #include <sys/pioctl.h>
  59 #include <sys/ptrace.h>
  60 #include <sys/kdb.h>
  61 #include <sys/kernel.h>
  62 #include <sys/ktr.h>
  63 #include <sys/lock.h>
  64 #include <sys/mutex.h>
  65 #include <sys/resourcevar.h>
  66 #include <sys/signalvar.h>
  67 #include <sys/syscall.h>
  68 #include <sys/sysctl.h>
  69 #include <sys/sysent.h>
  70 #include <sys/uio.h>
  71 #include <sys/vmmeter.h>
  72 #ifdef HWPMC_HOOKS
  73 #include <sys/pmckern.h>
  74 PMC_SOFT_DEFINE( , , page_fault, all);
  75 PMC_SOFT_DEFINE( , , page_fault, read);
  76 PMC_SOFT_DEFINE( , , page_fault, write);
  77 #endif
  78
  79 #include <vm/vm.h>
  80 #include <vm/vm_param.h>
  81 #include <vm/pmap.h>
  82 #include <vm/vm_kern.h>
  83 #include <vm/vm_map.h>
  84 #include <vm/vm_page.h>
  85 #include <vm/vm_extern.h>
  86
  87 #include <machine/cpu.h>
  88 #include <machine/intr_machdep.h>
  89 #include <x86/mca.h>
  90 #include <machine/md_var.h>
  91 #include <machine/pcb.h>
  92 #ifdef SMP
  93 #include <machine/smp.h>
  94 #endif
  95 #include <machine/tss.h>
  96
  97 #ifdef KDTRACE_HOOKS
  98 #include <sys/dtrace_bsd.h>
  99 #endif
 100
 101 extern void trap(struct trapframe *frame);
 102 extern void syscall(struct trapframe *frame);
 103 void dblfault_handler(struct trapframe *frame);
 104
 105 static int trap_pfault(struct trapframe *, int);
 106 static void trap_fatal(struct trapframe *, vm_offset_t);
 107
 108 #define MAX_TRAP_MSG            32
 109 static char *trap_msg[] = {
 110         "",                                     /*  0 unused */
 111         "privileged instruction fault",         /*  1 T_PRIVINFLT */
 112         "",                                     /*  2 unused */
 113         "breakpoint instruction fault",         /*  3 T_BPTFLT */
 114         "",                                     /*  4 unused */
 115         "",                                     /*  5 unused */
 116         "arithmetic trap",                      /*  6 T_ARITHTRAP */
 117         "",                                     /*  7 unused */
 118         "",                                     /*  8 unused */
 119         "general protection fault",             /*  9 T_PROTFLT */
 120         "trace trap",                           /* 10 T_TRCTRAP */
 121         "",                                     /* 11 unused */
 122         "page fault",                           /* 12 T_PAGEFLT */
 123         "",                                     /* 13 unused */
 124         "alignment fault",                      /* 14 T_ALIGNFLT */
 125         "",                                     /* 15 unused */
 126         "",                                     /* 16 unused */
 127         "",                                     /* 17 unused */
 128         "integer divide fault",                 /* 18 T_DIVIDE */
 129         "non-maskable interrupt trap",          /* 19 T_NMI */
 130         "overflow trap",                        /* 20 T_OFLOW */
 131         "FPU bounds check fault",               /* 21 T_BOUND */
 132         "FPU device not available",             /* 22 T_DNA */
 133         "double fault",                         /* 23 T_DOUBLEFLT */
 134         "FPU operand fetch fault",              /* 24 T_FPOPFLT */
 135         "invalid TSS fault",                    /* 25 T_TSSFLT */
 136         "segment not present fault",            /* 26 T_SEGNPFLT */
 137         "stack fault",                          /* 27 T_STKFLT */
 138         "machine check trap",                   /* 28 T_MCHK */
 139         "SIMD floating-point exception",        /* 29 T_XMMFLT */
 140         "reserved (unknown) fault",             /* 30 T_RESERVED */
 141         "",                                     /* 31 unused (reserved) */
 142         "DTrace pid return trap",               /* 32 T_DTRACE_RET */
 143 };
 144
 145 #ifdef KDB
 146 static int kdb_on_nmi = 1;
 147 SYSCTL_INT(_machdep, OID_AUTO, kdb_on_nmi, CTLFLAG_RW,
 148         &kdb_on_nmi, 0, "Go to KDB on NMI");
 149 TUNABLE_INT("machdep.kdb_on_nmi", &kdb_on_nmi);
 150 #endif
 151 static int panic_on_nmi = 1;
 152 SYSCTL_INT(_machdep, OID_AUTO, panic_on_nmi, CTLFLAG_RW,
 153         &panic_on_nmi, 0, "Panic on NMI");
 154 TUNABLE_INT("machdep.panic_on_nmi", &panic_on_nmi);
 155 static int prot_fault_translation;
 156 SYSCTL_INT(_machdep, OID_AUTO, prot_fault_translation, CTLFLAG_RW,
 157     &prot_fault_translation, 0,
 158     "Select signal to deliver on protection fault");
 159 static int uprintf_signal;
 160 SYSCTL_INT(_machdep, OID_AUTO, uprintf_signal, CTLFLAG_RW,
 161     &uprintf_signal, 0,
 162     "Print debugging information on trap signal to ctty");
 163
 164 /*
 165  * Exception, fault, and trap interface to the FreeBSD kernel.
 166  * This common code is called from assembly language IDT gate entry
 167  * routines that prepare a suitable stack frame, and restore this
 168  * frame after the exception has been processed.
 169  */
 170
 171 void
 172 trap(struct trapframe *frame)
 173 {
 174 #ifdef KDTRACE_HOOKS
 175         struct reg regs;
 176 #endif
 177         struct thread *td = curthread;
 178         struct proc *p = td->td_proc;
 179         int i = 0, ucode = 0, code;
 180         u_int type;
 181         register_t addr = 0;
 182         ksiginfo_t ksi;
 183
 184         PCPU_INC(cnt.v_trap);
 185         type = frame->tf_trapno;
 186
 187 #ifdef SMP
 188         /* Handler for NMI IPIs used for stopping CPUs. */
 189         if (type == T_NMI) {
 190                  if (ipi_nmi_handler() == 0)
 191                            goto out;
 192         }
 193 #endif /* SMP */
 194
 195 #ifdef KDB
 196         if (kdb_active) {
 197                 kdb_reenter();
 198                 goto out;
 199         }
 200 #endif
 201
 202         if (type == T_RESERVED) {
 203                 trap_fatal(frame, 0);
 204                 goto out;
 205         }
 206
 207 #ifdef  HWPMC_HOOKS
 208         /*
 209          * CPU PMCs interrupt using an NMI.  If the PMC module is
 210          * active, pass the 'rip' value to the PMC module's interrupt
 211          * handler.  A return value of '1' from the handler means that
 212          * the NMI was handled by it and we can return immediately.
 213          */
 214         if (type == T_NMI && pmc_intr &&
 215             (*pmc_intr)(PCPU_GET(cpuid), frame))
 216                 goto out;
 217 #endif
 218
 219         if (type == T_MCHK) {
 220                 mca_intr();
 221                 goto out;
 222         }
 223
 224 #ifdef KDTRACE_HOOKS
 225         /*
 226          * A trap can occur while DTrace executes a probe. Before
 227          * executing the probe, DTrace blocks re-scheduling and sets
 228          * a flag in its per-cpu flags to indicate that it doesn't
 229          * want to fault. On returning from the probe, the no-fault
 230          * flag is cleared and finally re-scheduling is enabled.
 231          */
 232         if (dtrace_trap_func != NULL && (*dtrace_trap_func)(frame, type))
 233                 goto out;
 234 #endif
 235
 236         if ((frame->tf_rflags & PSL_I) == 0) {
 237                 /*
 238                  * Buggy application or kernel code has disabled
 239                  * interrupts and then trapped.  Enabling interrupts
 240                  * now is wrong, but it is better than running with
 241                  * interrupts disabled until they are accidentally
 242                  * enabled later.
 243                  */
 244                 if (ISPL(frame->tf_cs) == SEL_UPL)
 245                         uprintf(
 246                             "pid %ld (%s): trap %d with interrupts disabled\n",
 247                             (long)curproc->p_pid, curthread->td_name, type);
 248                 else if (type != T_NMI && type != T_BPTFLT &&
 249                     type != T_TRCTRAP) {
 250                         /*
 251                          * XXX not quite right, since this may be for a
 252                          * multiple fault in user mode.
 253                          */
 254                         printf("kernel trap %d with interrupts disabled\n",
 255                             type);
 256
 257                         /*
 258                          * We shouldn't enable interrupts while holding a
 259                          * spin lock.
 260                          */
 261                         if (td->td_md.md_spinlock_count == 0)
 262                                 enable_intr();
 263                 }
 264         }
 265
 266         code = frame->tf_err;
 267
 268         if (ISPL(frame->tf_cs) == SEL_UPL) {
 269                 /* user trap */
 270
 271                 td->td_pticks = 0;
 272                 td->td_frame = frame;
 273                 addr = frame->tf_rip;
 274                 if (td->td_ucred != p->p_ucred)
 275                         cred_update_thread(td);
 276
 277                 switch (type) {
 278                 case T_PRIVINFLT:       /* privileged instruction fault */
 279                         i = SIGILL;
 280                         ucode = ILL_PRVOPC;
 281                         break;
 282
 283                 case T_BPTFLT:          /* bpt instruction fault */
 284                 case T_TRCTRAP:         /* trace trap */
 285                         enable_intr();
 286 #ifdef KDTRACE_HOOKS
 287                         if (type == T_BPTFLT) {
 288                                 fill_frame_regs(frame, &regs);
 289                                 if (dtrace_pid_probe_ptr != NULL &&
 290                                     dtrace_pid_probe_ptr(&regs) == 0)
 291                                         goto out;
 292                         }
 293 #endif
 294                         frame->tf_rflags &= ~PSL_T;
 295                         i = SIGTRAP;
 296                         ucode = (type == T_TRCTRAP ? TRAP_TRACE : TRAP_BRKPT);
 297                         break;
 298
 299                 case T_ARITHTRAP:       /* arithmetic trap */
 300                         ucode = fputrap_x87();
 301                         if (ucode == -1)
 302                                 goto userout;
 303                         i = SIGFPE;
 304                         break;
 305
 306                 case T_PROTFLT:         /* general protection fault */
 307                         i = SIGBUS;
 308                         ucode = BUS_OBJERR;
 309                         break;
 310                 case T_STKFLT:          /* stack fault */
 311                 case T_SEGNPFLT:        /* segment not present fault */
 312                         i = SIGBUS;
 313                         ucode = BUS_ADRERR;
 314                         break;
 315                 case T_TSSFLT:          /* invalid TSS fault */
 316                         i = SIGBUS;
 317                         ucode = BUS_OBJERR;
 318                         break;
 319                 case T_ALIGNFLT:
 320                         i = SIGBUS;
 321                         ucode = BUS_ADRALN;
 322                         break;
 323                 case T_DOUBLEFLT:       /* double fault */
 324                 default:
 325                         i = SIGBUS;
 326                         ucode = BUS_OBJERR;
 327                         break;
 328
 329                 case T_PAGEFLT:         /* page fault */
 330                         /*
 331                          * Emulator can take care about this trap?
 332                          */
 333                         if (*p->p_sysent->sv_trap != NULL &&
 334                             (*p->p_sysent->sv_trap)(td) == 0)
 335                                 goto userout;
 336
 337                         addr = frame->tf_addr;
 338                         i = trap_pfault(frame, TRUE);
 339                         if (i == -1)
 340                                 goto userout;
 341                         if (i == 0)
 342                                 goto user;
 343
 344                         if (i == SIGSEGV)
 345                                 ucode = SEGV_MAPERR;
 346                         else {
 347                                 if (prot_fault_translation == 0) {
 348                                         /*
 349                                          * Autodetect.
 350                                          * This check also covers the images
 351                                          * without the ABI-tag ELF note.
 352                                          */
 353                                         if (SV_CURPROC_ABI() == SV_ABI_FREEBSD
 354                                             && p->p_osrel >= P_OSREL_SIGSEGV) {
 355                                                 i = SIGSEGV;
 356                                                 ucode = SEGV_ACCERR;
 357                                         } else {
 358                                                 i = SIGBUS;
 359                                                 ucode = BUS_PAGE_FAULT;
 360                                         }
 361                                 } else if (prot_fault_translation == 1) {
 362                                         /*
 363                                          * Always compat mode.
 364                                          */
 365                                         i = SIGBUS;
 366                                         ucode = BUS_PAGE_FAULT;
 367                                 } else {
 368                                         /*
 369                                          * Always SIGSEGV mode.
 370                                          */
 371                                         i = SIGSEGV;
 372                                         ucode = SEGV_ACCERR;
 373                                 }
 374                         }
 375                         break;
 376
 377                 case T_DIVIDE:          /* integer divide fault */
 378                         ucode = FPE_INTDIV;
 379                         i = SIGFPE;
 380                         break;
 381
 382 #ifdef DEV_ISA
 383                 case T_NMI:
 384                         /* machine/parity/power fail/"kitchen sink" faults */
 385                         if (isa_nmi(code) == 0) {
 386 #ifdef KDB
 387                                 /*
 388                                  * NMI can be hooked up to a pushbutton
 389                                  * for debugging.
 390                                  */
 391                                 if (kdb_on_nmi) {
 392                                         printf ("NMI ... going to debugger\n");
 393                                         kdb_trap(type, 0, frame);
 394                                 }
 395 #endif /* KDB */
 396                                 goto userout;
 397                         } else if (panic_on_nmi)
 398                                 panic("NMI indicates hardware failure");
 399                         break;
 400 #endif /* DEV_ISA */
 401
 402                 case T_OFLOW:           /* integer overflow fault */
 403                         ucode = FPE_INTOVF;
 404                         i = SIGFPE;
 405                         break;
 406
 407                 case T_BOUND:           /* bounds check fault */
 408                         ucode = FPE_FLTSUB;
 409                         i = SIGFPE;
 410                         break;
 411
 412                 case T_DNA:
 413                         /* transparent fault (due to context switch "late") */
 414                         KASSERT(PCB_USER_FPU(td->td_pcb),
 415                             ("kernel FPU ctx has leaked"));
 416                         fpudna();
 417                         goto userout;
 418
 419                 case T_FPOPFLT:         /* FPU operand fetch fault */
 420                         ucode = ILL_COPROC;
 421                         i = SIGILL;
 422                         break;
 423
 424                 case T_XMMFLT:          /* SIMD floating-point exception */
 425                         ucode = fputrap_sse();
 426                         if (ucode == -1)
 427                                 goto userout;
 428                         i = SIGFPE;
 429                         break;
 430 #ifdef KDTRACE_HOOKS
 431                 case T_DTRACE_RET:
 432                         enable_intr();
 433                         fill_frame_regs(frame, &regs);
 434                         if (dtrace_return_probe_ptr != NULL &&
 435                             dtrace_return_probe_ptr(&regs) == 0)
 436                                 goto out;
 437                         break;
 438 #endif
 439                 }
 440         } else {
 441                 /* kernel trap */
 442
 443                 KASSERT(cold || td->td_ucred != NULL,
 444                     ("kernel trap doesn't have ucred"));
 445                 switch (type) {
 446                 case T_PAGEFLT:                 /* page fault */
 447                         (void) trap_pfault(frame, FALSE);
 448                         goto out;
 449
 450                 case T_DNA:
 451                         KASSERT(!PCB_USER_FPU(td->td_pcb),
 452                             ("Unregistered use of FPU in kernel"));
 453                         fpudna();
 454                         goto out;
 455
 456                 case T_ARITHTRAP:       /* arithmetic trap */
 457                 case T_XMMFLT:          /* SIMD floating-point exception */
 458                 case T_FPOPFLT:         /* FPU operand fetch fault */
 459                         /*
 460                          * For now, supporting kernel handler
 461                          * registration for FPU traps is overkill.
 462                          */
 463                         trap_fatal(frame, 0);
 464                         goto out;
 465
 466                 case T_STKFLT:          /* stack fault */
 467                 case T_PROTFLT:         /* general protection fault */
 468                 case T_SEGNPFLT:        /* segment not present fault */
 469                         if (td->td_intr_nesting_level != 0)
 470                                 break;
 471
 472                         /*
 473                          * Invalid segment selectors and out of bounds
 474                          * %rip's and %rsp's can be set up in user mode.
 475                          * This causes a fault in kernel mode when the
 476                          * kernel tries to return to user mode.  We want
 477                          * to get this fault so that we can fix the
 478                          * problem here and not have to check all the
 479                          * selectors and pointers when the user changes
 480                          * them.
 481                          */
 482                         if (frame->tf_rip == (long)doreti_iret) {
 483                                 frame->tf_rip = (long)doreti_iret_fault;
 484                                 goto out;
 485                         }
 486                         if (frame->tf_rip == (long)ld_ds) {
 487                                 frame->tf_rip = (long)ds_load_fault;
 488                                 goto out;
 489                         }
 490                         if (frame->tf_rip == (long)ld_es) {
 491                                 frame->tf_rip = (long)es_load_fault;
 492                                 goto out;
 493                         }
 494                         if (frame->tf_rip == (long)ld_fs) {
 495                                 frame->tf_rip = (long)fs_load_fault;
 496                                 goto out;
 497                         }
 498                         if (frame->tf_rip == (long)ld_gs) {
 499                                 frame->tf_rip = (long)gs_load_fault;
 500                                 goto out;
 501                         }
 502                         if (frame->tf_rip == (long)ld_gsbase) {
 503                                 frame->tf_rip = (long)gsbase_load_fault;
 504                                 goto out;
 505                         }
 506                         if (frame->tf_rip == (long)ld_fsbase) {
 507                                 frame->tf_rip = (long)fsbase_load_fault;
 508                                 goto out;
 509                         }
 510                         if (curpcb->pcb_onfault != NULL) {
 511                                 frame->tf_rip = (long)curpcb->pcb_onfault;
 512                                 goto out;
 513                         }
 514                         break;
 515
 516                 case T_TSSFLT:
 517                         /*
 518                          * PSL_NT can be set in user mode and isn't cleared
 519                          * automatically when the kernel is entered.  This
 520                          * causes a TSS fault when the kernel attempts to
 521                          * `iret' because the TSS link is uninitialized.  We
 522                          * want to get this fault so that we can fix the
 523                          * problem here and not every time the kernel is
 524                          * entered.
 525                          */
 526                         if (frame->tf_rflags & PSL_NT) {
 527                                 frame->tf_rflags &= ~PSL_NT;
 528                                 goto out;
 529                         }
 530                         break;
 531
 532                 case T_TRCTRAP:  /* trace trap */
 533                         /*
 534                          * Ignore debug register trace traps due to
 535                          * accesses in the user's address space, which
 536                          * can happen under several conditions such as
 537                          * if a user sets a watchpoint on a buffer and
 538                          * then passes that buffer to a system call.
 539                          * We still want to get TRCTRAPS for addresses
 540                          * in kernel space because that is useful when
 541                          * debugging the kernel.
 542                          */
 543                         if (user_dbreg_trap()) {
 544                                 /*
 545                                  * Reset breakpoint bits because the
 546                                  * processor doesn't
 547                                  */
 548                                 /* XXX check upper bits here */
 549                                 load_dr6(rdr6() & 0xfffffff0);
 550                                 goto out;
 551                         }
 552                         /*
 553                          * FALLTHROUGH (TRCTRAP kernel mode, kernel address)
 554                          */
 555                 case T_BPTFLT:
 556                         /*
 557                          * If KDB is enabled, let it handle the debugger trap.
 558                          * Otherwise, debugger traps "can't happen".
 559                          */
 560 #ifdef KDB
 561                         if (kdb_trap(type, 0, frame))
 562                                 goto out;
 563 #endif
 564                         break;
 565
 566 #ifdef DEV_ISA
 567                 case T_NMI:
 568                         /* machine/parity/power fail/"kitchen sink" faults */
 569                         if (isa_nmi(code) == 0) {
 570 #ifdef KDB
 571                                 /*
 572                                  * NMI can be hooked up to a pushbutton
 573                                  * for debugging.
 574                                  */
 575                                 if (kdb_on_nmi) {
 576                                         printf ("NMI ... going to debugger\n");
 577                                         kdb_trap(type, 0, frame);
 578                                 }
 579 #endif /* KDB */
 580                                 goto out;
 581                         } else if (panic_on_nmi == 0)
 582                                 goto out;
 583                         /* FALLTHROUGH */
 584 #endif /* DEV_ISA */
 585                 }
 586
 587                 trap_fatal(frame, 0);
 588                 goto out;
 589         }
 590
 591         /* Translate fault for emulators (e.g. Linux) */
 592         if (*p->p_sysent->sv_transtrap)
 593                 i = (*p->p_sysent->sv_transtrap)(i, type);
 594
 595         ksiginfo_init_trap(&ksi);
 596         ksi.ksi_signo = i;
 597         ksi.ksi_code = ucode;
 598         ksi.ksi_trapno = type;
 599         ksi.ksi_addr = (void *)addr;
 600         if (uprintf_signal) {
 601                 uprintf("pid %d comm %s: signal %d err %lx code %d type %d "
 602                     "addr 0x%lx rsp 0x%lx rip 0x%lx "
 603                     "<%02x %02x %02x %02x %02x %02x %02x %02x>\n",
 604                     p->p_pid, p->p_comm, i, frame->tf_err, ucode, type, addr,
 605                     frame->tf_rsp, frame->tf_rip,
 606                     fubyte((void *)(frame->tf_rip + 0)),
 607                     fubyte((void *)(frame->tf_rip + 1)),
 608                     fubyte((void *)(frame->tf_rip + 2)),
 609                     fubyte((void *)(frame->tf_rip + 3)),
 610                     fubyte((void *)(frame->tf_rip + 4)),
 611                     fubyte((void *)(frame->tf_rip + 5)),
 612                     fubyte((void *)(frame->tf_rip + 6)),
 613                     fubyte((void *)(frame->tf_rip + 7)));
 614         }
 615         KASSERT((read_rflags() & PSL_I) != 0, ("interrupts disabled"));
 616         trapsignal(td, &ksi);
 617
 618 user:
 619         userret(td, frame);
 620         KASSERT(PCB_USER_FPU(td->td_pcb),
 621             ("Return from trap with kernel FPU ctx leaked"));
 622 userout:
 623 out:
 624         return;
 625 }
 626
 627 static int
 628 trap_pfault(frame, usermode)
 629         struct trapframe *frame;
 630         int usermode;
 631 {
 632         vm_offset_t va;
 633         struct vmspace *vm;
 634         vm_map_t map;
 635         int rv = 0;
 636         vm_prot_t ftype;
 637         struct thread *td = curthread;
 638         struct proc *p = td->td_proc;
 639         vm_offset_t eva = frame->tf_addr;
 640
 641         if (__predict_false((td->td_pflags & TDP_NOFAULTING) != 0)) {
 642                 /*
 643                  * Due to both processor errata and lazy TLB invalidation when
 644                  * access restrictions are removed from virtual pages, memory
 645                  * accesses that are allowed by the physical mapping layer may
 646                  * nonetheless cause one spurious page fault per virtual page.
 647                  * When the thread is executing a "no faulting" section that
 648                  * is bracketed by vm_fault_{disable,enable}_pagefaults(),
 649                  * every page fault is treated as a spurious page fault,
 650                  * unless it accesses the same virtual address as the most
 651                  * recent page fault within the same "no faulting" section.
 652                  */
 653                 if (td->td_md.md_spurflt_addr != eva ||
 654                     (td->td_pflags & TDP_RESETSPUR) != 0) {
 655                         /*
 656                          * Do nothing to the TLB.  A stale TLB entry is
 657                          * flushed automatically by a page fault.
 658                          */
 659                         td->td_md.md_spurflt_addr = eva;
 660                         td->td_pflags &= ~TDP_RESETSPUR;
 661                         return (0);
 662                 }
 663         } else {
 664                 /*
 665                  * If we get a page fault while in a critical section, then
 666                  * it is most likely a fatal kernel page fault.  The kernel
 667                  * is already going to panic trying to get a sleep lock to
 668                  * do the VM lookup, so just consider it a fatal trap so the
 669                  * kernel can print out a useful trap message and even get
 670                  * to the debugger.
 671                  *
 672                  * If we get a page fault while holding a non-sleepable
 673                  * lock, then it is most likely a fatal kernel page fault.
 674                  * If WITNESS is enabled, then it's going to whine about
 675                  * bogus LORs with various VM locks, so just skip to the
 676                  * fatal trap handling directly.
 677                  */
 678                 if (td->td_critnest != 0 ||
 679                     WITNESS_CHECK(WARN_SLEEPOK | WARN_GIANTOK, NULL,
 680                     "Kernel page fault") != 0) {
 681                         trap_fatal(frame, eva);
 682                         return (-1);
 683                 }
 684         }
 685         va = trunc_page(eva);
 686         if (va >= VM_MIN_KERNEL_ADDRESS) {
 687                 /*
 688                  * Don't allow user-mode faults in kernel address space.
 689                  */
 690                 if (usermode)
 691                         goto nogo;
 692
 693                 map = kernel_map;
 694         } else {
 695                 /*
 696                  * This is a fault on non-kernel virtual memory.  If either
 697                  * p or p->p_vmspace is NULL, then the fault is fatal.
 698                  */
 699                 if (p == NULL || (vm = p->p_vmspace) == NULL)
 700                         goto nogo;
 701
 702                 map = &vm->vm_map;
 703
 704                 /*
 705                  * When accessing a usermode address, kernel must be
 706                  * ready to accept the page fault, and provide a
 707                  * handling routine.  Since accessing the address
 708                  * without the handler is a bug, do not try to handle
 709                  * it normally, and panic immediately.
 710                  */
 711                 if (!usermode && (td->td_intr_nesting_level != 0 ||
 712                     curpcb->pcb_onfault == NULL)) {
 713                         trap_fatal(frame, eva);
 714                         return (-1);
 715                 }
 716         }
 717
 718         /*
 719          * If the trap was caused by errant bits in the PTE then panic.
 720          */
 721         if (frame->tf_err & PGEX_RSV) {
 722                 trap_fatal(frame, eva);
 723                 return (-1);
 724         }
 725
 726         /*
 727          * PGEX_I is defined only if the execute disable bit capability is
 728          * supported and enabled.
 729          */
 730         if (frame->tf_err & PGEX_W)
 731                 ftype = VM_PROT_WRITE;
 732         else if ((frame->tf_err & PGEX_I) && pg_nx != 0)
 733                 ftype = VM_PROT_EXECUTE;
 734         else
 735                 ftype = VM_PROT_READ;
 736
 737         if (map != kernel_map) {
 738                 /*
 739                  * Keep swapout from messing with us during this
 740                  *      critical time.
 741                  */
 742                 PROC_LOCK(p);
 743                 ++p->p_lock;
 744                 PROC_UNLOCK(p);
 745
 746                 /* Fault in the user page: */
 747                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 748
 749                 PROC_LOCK(p);
 750                 --p->p_lock;
 751                 PROC_UNLOCK(p);
 752         } else {
 753                 /*
 754                  * Don't have to worry about process locking or stacks in the
 755                  * kernel.
 756                  */
 757                 rv = vm_fault(map, va, ftype, VM_FAULT_NORMAL);
 758         }
 759         if (rv == KERN_SUCCESS) {
 760 #ifdef HWPMC_HOOKS
 761                 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
 762                         PMC_SOFT_CALL_TF( , , page_fault, all, frame);
 763                         if (ftype == VM_PROT_READ)
 764                                 PMC_SOFT_CALL_TF( , , page_fault, read,
 765                                     frame);
 766                         else
 767                                 PMC_SOFT_CALL_TF( , , page_fault, write,
 768                                     frame);
 769                 }
 770 #endif
 771                 return (0);
 772         }
 773 nogo:
 774         if (!usermode) {
 775                 if (td->td_intr_nesting_level == 0 &&
 776                     curpcb->pcb_onfault != NULL) {
 777                         frame->tf_rip = (long)curpcb->pcb_onfault;
 778                         return (0);
 779                 }
 780                 trap_fatal(frame, eva);
 781                 return (-1);
 782         }
 783         return ((rv == KERN_PROTECTION_FAILURE) ? SIGBUS : SIGSEGV);
 784 }
 785
 786 static void
 787 trap_fatal(frame, eva)
 788         struct trapframe *frame;
 789         vm_offset_t eva;
 790 {
 791         int code, ss;
 792         u_int type;
 793         long esp;
 794         struct soft_segment_descriptor softseg;
 795         char *msg;
 796
 797         code = frame->tf_err;
 798         type = frame->tf_trapno;
 799         sdtossd(&gdt[NGDT * PCPU_GET(cpuid) + IDXSEL(frame->tf_cs & 0xffff)],
 800             &softseg);
 801
 802         if (type <= MAX_TRAP_MSG)
 803                 msg = trap_msg[type];
 804         else
 805                 msg = "UNKNOWN";
 806         printf("\n\nFatal trap %d: %s while in %s mode\n", type, msg,
 807             ISPL(frame->tf_cs) == SEL_UPL ? "user" : "kernel");
 808 #ifdef SMP
 809         /* two separate prints in case of a trap on an unmapped page */
 810         printf("cpuid = %d; ", PCPU_GET(cpuid));
 811         printf("apic id = %02x\n", PCPU_GET(apic_id));
 812 #endif
 813         if (type == T_PAGEFLT) {
 814                 printf("fault virtual address   = 0x%lx\n", eva);
 815                 printf("fault code              = %s %s %s%s, %s\n",
 816                         code & PGEX_U ? "user" : "supervisor",
 817                         code & PGEX_W ? "write" : "read",
 818                         code & PGEX_I ? "instruction" : "data",
 819                         code & PGEX_RSV ? " rsv" : "",
 820                         code & PGEX_P ? "protection violation" : "page not present");
 821         }
 822         printf("instruction pointer     = 0x%lx:0x%lx\n",
 823                frame->tf_cs & 0xffff, frame->tf_rip);
 824         if (ISPL(frame->tf_cs) == SEL_UPL) {
 825                 ss = frame->tf_ss & 0xffff;
 826                 esp = frame->tf_rsp;
 827         } else {
 828                 ss = GSEL(GDATA_SEL, SEL_KPL);
 829                 esp = (long)&frame->tf_rsp;
 830         }
 831         printf("stack pointer           = 0x%x:0x%lx\n", ss, esp);
 832         printf("frame pointer           = 0x%x:0x%lx\n", ss, frame->tf_rbp);
 833         printf("code segment            = base 0x%lx, limit 0x%lx, type 0x%x\n",
 834                softseg.ssd_base, softseg.ssd_limit, softseg.ssd_type);
 835         printf("                        = DPL %d, pres %d, long %d, def32 %d, gran %d\n",
 836                softseg.ssd_dpl, softseg.ssd_p, softseg.ssd_long, softseg.ssd_def32,
 837                softseg.ssd_gran);
 838         printf("processor eflags        = ");
 839         if (frame->tf_rflags & PSL_T)
 840                 printf("trace trap, ");
 841         if (frame->tf_rflags & PSL_I)
 842                 printf("interrupt enabled, ");
 843         if (frame->tf_rflags & PSL_NT)
 844                 printf("nested task, ");
 845         if (frame->tf_rflags & PSL_RF)
 846                 printf("resume, ");
 847         printf("IOPL = %ld\n", (frame->tf_rflags & PSL_IOPL) >> 12);
 848         printf("current process         = %d (%s)\n",
 849             curproc->p_pid, curthread->td_name);
 850
 851 #ifdef KDB
 852         if (debugger_on_panic || kdb_active)
 853                 if (kdb_trap(type, 0, frame))
 854                         return;
 855 #endif
 856         printf("trap number             = %d\n", type);
 857         if (type <= MAX_TRAP_MSG)
 858                 panic("%s", trap_msg[type]);
 859         else
 860                 panic("unknown/reserved trap");
 861 }
 862
 863 /*
 864  * Double fault handler. Called when a fault occurs while writing
 865  * a frame for a trap/exception onto the stack. This usually occurs
 866  * when the stack overflows (such is the case with infinite recursion,
 867  * for example).
 868  */
 869 void
 870 dblfault_handler(struct trapframe *frame)
 871 {
 872 #ifdef KDTRACE_HOOKS
 873         if (dtrace_doubletrap_func != NULL)
 874                 (*dtrace_doubletrap_func)();
 875 #endif
 876         printf("\nFatal double fault\n");
 877         printf("rip = 0x%lx\n", frame->tf_rip);
 878         printf("rsp = 0x%lx\n", frame->tf_rsp);
 879         printf("rbp = 0x%lx\n", frame->tf_rbp);
 880 #ifdef SMP
 881         /* two separate prints in case of a trap on an unmapped page */
 882         printf("cpuid = %d; ", PCPU_GET(cpuid));
 883         printf("apic id = %02x\n", PCPU_GET(apic_id));
 884 #endif
 885         panic("double fault");
 886 }
 887
 888 int
 889 cpu_fetch_syscall_args(struct thread *td, struct syscall_args *sa)
 890 {
 891         struct proc *p;
 892         struct trapframe *frame;
 893         register_t *argp;
 894         caddr_t params;
 895         int reg, regcnt, error;
 896
 897         p = td->td_proc;
 898         frame = td->td_frame;
 899         reg = 0;
 900         regcnt = 6;
 901
 902         params = (caddr_t)frame->tf_rsp + sizeof(register_t);
 903         sa->code = frame->tf_rax;
 904
 905         if (sa->code == SYS_syscall || sa->code == SYS___syscall) {
 906                 sa->code = frame->tf_rdi;
 907                 reg++;
 908                 regcnt--;
 909         }
 910         if (p->p_sysent->sv_mask)
 911                 sa->code &= p->p_sysent->sv_mask;
 912
 913         if (sa->code >= p->p_sysent->sv_size)
 914                 sa->callp = &p->p_sysent->sv_table[0];
 915         else
 916                 sa->callp = &p->p_sysent->sv_table[sa->code];
 917
 918         sa->narg = sa->callp->sy_narg;
 919         KASSERT(sa->narg <= sizeof(sa->args) / sizeof(sa->args[0]),
 920             ("Too many syscall arguments!"));
 921         error = 0;
 922         argp = &frame->tf_rdi;
 923         argp += reg;
 924         bcopy(argp, sa->args, sizeof(sa->args[0]) * regcnt);
 925         if (sa->narg > regcnt) {
 926                 KASSERT(params != NULL, ("copyin args with no params!"));
 927                 error = copyin(params, &sa->args[regcnt],
 928                     (sa->narg - regcnt) * sizeof(sa->args[0]));
 929         }
 930
 931         if (error == 0) {
 932                 td->td_retval[0] = 0;
 933                 td->td_retval[1] = frame->tf_rdx;
 934         }
 935
 936         return (error);
 937 }
 938
 939 #include "../../kern/subr_syscall.c"
 940
 941 /*
 942  * System call handler for native binaries.  The trap frame is already
 943  * set up by the assembler trampoline and a pointer to it is saved in
 944  * td_frame.
 945  */
 946 void
 947 amd64_syscall(struct thread *td, int traced)
 948 {
 949         struct syscall_args sa;
 950         int error;
 951         ksiginfo_t ksi;
 952
 953 #ifdef DIAGNOSTIC
 954         if (ISPL(td->td_frame->tf_cs) != SEL_UPL) {
 955                 panic("syscall");
 956                 /* NOT REACHED */
 957         }
 958 #endif
 959         error = syscallenter(td, &sa);
 960
 961         /*
 962          * Traced syscall.
 963          */
 964         if (__predict_false(traced)) {
 965                 td->td_frame->tf_rflags &= ~PSL_T;
 966                 ksiginfo_init_trap(&ksi);
 967                 ksi.ksi_signo = SIGTRAP;
 968                 ksi.ksi_code = TRAP_TRACE;
 969                 ksi.ksi_addr = (void *)td->td_frame->tf_rip;
 970                 trapsignal(td, &ksi);
 971         }
 972
 973         KASSERT(PCB_USER_FPU(td->td_pcb),
 974             ("System call %s returing with kernel FPU ctx leaked",
 975              syscallname(td->td_proc, sa.code)));
 976         KASSERT(td->td_pcb->pcb_save == get_pcb_user_save_td(td),
 977             ("System call %s returning with mangled pcb_save",
 978              syscallname(td->td_proc, sa.code)));
 979
 980         syscallret(td, error, &sa);
 981
 982         /*
 983          * If the user-supplied value of %rip is not a canonical
 984          * address, then some CPUs will trigger a ring 0 #GP during
 985          * the sysret instruction.  However, the fault handler would
 986          * execute in ring 0 with the user's %gs and %rsp which would
 987          * not be safe.  Instead, use the full return path which
 988          * catches the problem safely.
 989          */
 990         if (td->td_frame->tf_rip >= VM_MAXUSER_ADDRESS)
 991                 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
 992 }